https://devblogs.nvidia.com/parallelforall/inside-pascal/
>28nm
>3072 shaders
>1114 MHz
>7.0 TFLOPs FP32
>250W TDP
>16nm
>3584 shaders
>1480 Mhz
>10.6TFLOPs FP32
>new memory
>new bus
>300W TDP
Embarrassing.
Extremely embarrassing.
Nvidia, what the fuck are you doing? Surely you HAD money for a new architecture instead of bolting HBM and FP64 to Maxwell.
What the fuck? SERIOUSLY what the fuck?
No wonder they've been so tight lipped about it.
I hope for Nvidia's sakes that after they remove those FP64 units for consumer cards the TDP drops by 50% or this is gonna be a huge, Fermi-tier mess again.
Those fucking clocks though
Jewvidia literally dead.
>>53883756
>610mm2
>only 512 more shaders
waht fug
this is the big Pascal chip?
Like damn I expected 4096 at the least.
>>53883756
>Like previous Tesla GPUs, GP100 is composed of an array of Graphics Processing Clusters (GPCs), Streaming Multiprocessors (SMs), and memory controllers. GP100 achieves its colossal throughput by providing six GPCs, up to 60 SMs, and eight 512-bit memory controllers (4096 bits total). The Pascal architecture’s computational prowess is more than just brute force: it increases performance not only by adding more SMs than previous GPUs, but by making each SM more efficient. Each SM has 64 CUDA cores and four texture units, for a total of 3840 CUDA cores and 240 texture units.
even if it was like you said it it would still 30% faster.
Also a more direct comparison.
>GM200
28000 GFLOP/W
>GP100
35333 GFLOP/W
Barely a 30% increase in perf/watt after a full node and half node jump.
If Nvidia can't get better perf/watt from removing those FP64 units this generation is gonna be terrible for them.
>>53883840
>610mm2
>is this big Pascal?
No, it's junior.
Of course it's big Pascal, even Intel can't make bigger than 660mm2 dies.
So only their high end card is 16nm/HBM... when's polaris?
>>53883911
I know dipshit, it's a rhetorical question to express disbelief even though I know it be what it do.
>>53883919
Polaris in June.
Vega, the chip fighting this HBM2 Pascal dildo in Q1 2017, same time when GP100 releases.
I am a #AMDmissile now.
Total disaster!
>>53883756
Those numbers look terrible considering twice the transistor count, why is the performance/watt increase so small?
>>53883756
It's done Jewvidya is over Ncucks btfo etc.
>>53883756
>Surely you HAD money for a new architecture
They do, but they don't need it, even if AMD makes a far superior chip Nvidia will outsell them.
Happened with RV700 and Evergreen, will happen again.
>>53883937
>June
Haaaaaaaa... OK, more waiting.
I hope the only use this chip has is FP64, otherwise it's a complete joke.
I knew something was fishy as soon as 300W part was announced.
LOOK AT ALL THIS AMDKEK DAMAGE CONTROL
>>53884105
Specs on Nvidia site, not mine, you only need to know basic subtraction to see something's wrong.
Very disappointing
I am a #AMDmissile now.
i just hope i don't get laid off when i start my job at nvidia
A M D
M
D
>>53883756
>Embarrassing
>Extremely embarrassing
why?
>>53884194
Too low SPFP
This is what happens when you spend all your budget on marketing rather than actual progress.
>>53883756
Pascal
>FP32 Compute
10.6 TFLOPs
>FP64 Compute
5.3 TFLOPs!
Maxwell
>FP32 Compute
6.10 TFLOPs
>FP64 Compute
0.20 TFLOPs
Okay kid
>>53884273
>FP64
Lol, you mean the metric Nvidia didn't give a shit about for the last 3 years that is completely useless to anyone outside of HPC?
>>53884273
And that's relevant to the average consumer how?
>>53883927
dont lie you little shit
>>53884297
>>53884315
>Tesla P100
>TESLA
>>53884273
Even if you remove the FP64 you still got the small FP32 increase you dumbass.
Nvidiots BTFO.
>>53884334
>Pascal.
so nvidia fucked up, but rumors are still standing that they are going to release a card in may/june, and amd a month later.
and HBM2 is probably march/april 2017
>>53884336
>FP32 @6.10 TFLOPs to FP32 @10.6 TFLOPs
>+~40%
>small FP32 increase
>>53883756
>tfw bought a 970
>dont have any intention of ever going above 1440p, current display is 1050p
>>53883756
>>53883863
>>53883872
>>53884011
Well Im sad now.
I really hope AMD kicks ass then
>>53883800
Literally nothing that Maxwell can't already do comfortably on stock voltage. This is basically just an AMD-style factory overclock with a couple of add-ons to an existing architecture, i.e. "rebrands". Pascal is a lie and just a Maxwell die shrink. There'll be nothing interesting until Volta in late 2017/early 2018.
>>53884336
>6.10 to 10.6
>bad
Oh..fuck off Roy...
I'm an #AMDmissile now
>waiting to upgrade
>this happends
what now?
>>53883989
twice the transistor for that fp64?
>>53883911
doesnt it become physically impossible to go bigger than that? With the speed of light and the time of 1 clock? Anyone knows whats the theoretical maximum for a die size?
>>53884511
Wait for it to come out anyway, then get the current gen gear at a discount as they cut prices to make way for Pascuck cards
DX12 SUCKS, WE ARE WAITING TIL DX12 SUCKS LESS UNTIL WE BAKE IT INTO OUR SILICON
AMD ARE DUMB, PLS IGNORE BENCHMARKS SHOWING DX12 THEY ARE LIES
~40% increase in FP32, massive increase in FP64, much higher bandwidth, what were you expecting op? Seeing more than a 40% increase in FP32 in a new GPU architecture is unheard of. This is pretty run for the mill. This is basically Fermi 2.0 which is good because higher compute cards ala Hawaii tend to have longer lifetimes than shit like Kepler
It's a HPC FP64 accelerator with big profit margins
Everyone here including the faggot OP is small time thinkers that can't see the bigger picture
>>53883840
TSMC is using 20nm interconnects for their 16nm nodes, no wonder they can fit less
http://www.eetimes.com/document.asp?doc_id=1328866
>>53884590
Absolutely not of importance for the gaymen here.
So when is AMD releasing their shit for this generation?
I hope this isn't their only large die configuration and there is a die with the 1700+ FP64 cores replaced by FP32 cores.
If not, fuck, enjoy your 20% faster than 980Ti in games.
>>53884620
that settles it
even async is AMD ondly, im buying MAD this year
>>53884581
Depends on the metal layers involved, a interposer like Fiji can go up to 1000mm2, but these are rather simple.
The largest single chip was an Itanium many years ago, around 700mm2 something, so probably not much larger.
>>53884590
I wonder how much of that tflop increase is coming from increased clocks? People are running Titan Xes at 1550 even 1600 MHz with bios mods. 14 nm helps at low leakage but it seems there is only minor architecture change
Using Intel's terms, this looks more like a 'Tick' than a 'Tock'
https://rbt.asia/g/?task=search&ghost=yes&search_text=pascal+fp64+strapped
Ha
Ha
>>53884683
This! THIS!
This guy knows what's up, FP64 units are taking up too much diespace, hopefully there's a 600mm2 die without them, hopefully for all of us.
But I doubt it, those are some expensive as fuck masks and yields will be ugly, it's highly likely we'll get a CONSUMER version at 350-400mm2 with around 3000 FP32 shaders and around 12-13TFLOPS
>>53884749
Not surprised, Nvidia hasn't fared all to well doing new architectures and jumping nodes at the same time historically. Even Intel doesn't like doing that.
AMD doesn't give a fuck though.
Nvidia on suicidewatch
>>53883756
oh boy.
>>53883756
You ignored:
>SM count
>double-precision performance
>global memory bandwidth
Nice try.
Judging a GPU by single-precision FLOPS is stupid. Single-precision FLOPS has little to with real-world performance, particularly in the HPC market.
>>53885024
All the gaymen here don't care about that.
IT'S FERMI ALL OVER AGAIN
>>53883756
>28nm
all of my wat
>>53885024
No I didn't, read the last paragraph, twinklebutt.
There needs to be a die config with more FP32 and less FP64, because this card is ONLY useful for FP64.
>>53885038
Even for gaymen it's stupid to just go by FLOPS. Even games are often bandwidth-bound rather than compute-bound.
>>53885094
Nah. It's still higher-performance for FP32. Bandwidth usually matters more than FLOPS these days.
>nvidia release a card with low fp64 performance and good fp32 performance (maxwell)
>muh slow bitcoin, muh compute suck, fucking gayming card
>nvidia release a card with low fp32 performance and good fp64 performance (gp100)
>it's not the market, we need more gayming, a pro card lolololol
FUCK YOU /g/ GO BACK ON /v/, THIS IS A PRO CARD, NOT A GAYMING CARD.
>>53883756
I dug out an oldie just for this occasion.
>>53885177
Bandwidth only matters more at 4k+ and insane AA levels
>>53885146
Most bandwidth improvements like Nvlink are exclusive to Tesla. Not at all for gaymin.
>>53885187
If they cut down FP64 to 1/32 or /16 using this die they might actually get those numbers at around 170-180W
That would be quite decent, then I have nothing to complain about, other than the fact that a huge chunk of the die is used for fp64 units I'll never use.
>>53884581
the size limit is determined by the biggest masks the lithography process can use.
apparently the ~600mm^2 fiji's interposer (w/ 4x HBM modules) was pretty much the limit for TSMC or whoever made it.
610mm^2 plus HBM2 for Pascal makes this rumor more credible.
Guys, where are the ROPs? Is this a pure compute unit?
Ripe nvidiots
>w-wait for pascal, it'll be 10 times more powerful than current cards
Where are those people now? Waitfags BTFO
>>53885356
TMU's are still there, so no.
>>53885385
nice comment
>>53883756
>Polaris is GCN 1.3
>Polaris is 14nm
>Polaris has geometry processor for max performance in tessellation
>Polaris has more optimized asynchronous compute engines for max performance in DirectX 12
It's over, Nvidia is finished.
>>53885413
Nvidrones BTFO
>>53885177
>Bandwidth usually matters more than FLOPS these days.
that's because Nvidia's mania for tessellation literally throws bandwidth on the floor.
GPU use ROPs with 4x4 (AMD) or 4x2 (Nvidia) pixel blocks to exploit locality, but whenever you add 100s of thousands of 0.5-2 pixel triangles, you need to waste a full ROP cycle on the 80-95% of the unused pixels in the block.
extreme tessellation levels are more about sabotaging pixel throughput bandwidth than about saturating the tessellator unit itself.
>>53885413
I regret selling my GTX 970 now, but i will simply buy a Polaris anyway, screw Pascal, it will probably be more expensive anyway.
>people who don't understand things talking about things they don't understand
woooooooowwwww
>>53885285
you are a retard.
enterprise Pascal looks like a Hawaii knockoff, where every ALU can do two fp32 ops or one fp64 one per cycle.
Older Nvidia GPUs had dumber ALUs that were either pure fp32 or fp64, and the fp64 ones could do fp32 work by throwing away the extra precision.
While removing the "sharing" control logic could definitely reduce power, it's not gonna be >40% of the total power budget of the card, including memory, I/O, conversion losses, etc.
>>53885486
>Nvtards all round
WeeeeeeeeeeeW
>>53885417
where are you seeing that?
also, wouldn't their proprietary interconnect fabric controllers be pure wasted die space on a consumer card?
>all this AMD shilling
if you think this is disappointing, wait until you see what AMD brings out. they're replacing the 380x with a card that only draws 50w less than it for equal performance. how do i know? because i have the sample sitting on my bench right now. 400 series launches with 28nm rebrands as the flagship :/.
>>53885597
>Texture units
>>53885642
sorry, found it in the article text, but it does appear to be missing from OP's pic.
>>53885616
>my mums cousin who works as a janitor at Linus tech tips told me so.
>>53885673
kek, give it two weeks and /g/ will be flooded with disappointment from both sides. there is absolutely nothing about these new GPUs that makes me want to upgrade.
>>53885616
>wait
Wait for what, those were news of last week's MAD bankrupt&finished.
https://rbt.asia/g/thread/S53673068#p53674328
>>53885665
Open the nvidia site and check the 2nd image detailing the SM's, look at the bottom, there's texture units and texture cache.
>>53885616
Vega is going to be the real flagship, in early 2017, thanks to no HBM2 yet.
Polaris and first-gen Pascal will be GDDR5 only, not even GDDR5X, and it's not like it would be very smart to make a >400mm^2 16/14nm GPU that was starved for bandwidth (or be stuck in 4GB land).
GTX 1080/1090/whatever could be a real short-lived trainwreck if it's not delayed until GDDR5X availability in Fall/Winter.
>>53885734
>>53885616
Crazy Nvtards damage control.
>>53885792
Weird, there's still no ROPs
Are they coupled with the TMUs?
>>53883840
>waht fug
FP64 takes a lot of die space. 610mm^2 is most likely close to the rectangle limit of 16nm TSMC manufacturing process.
GP100 is clearly intended for scientific purposes only, this thing has 10 times more FP64 Tflops than Fury X and 25 times more than 980 Ti.
I doubt we ever see this GPU in consumer space.
>>53883756
Always remember that 1 Nvidia TFLOP is 30% faster than an AMD TFLOP because of drivers
>>53885871
Mad nvdrone shilling
>>53885871
Always remember that you were born more unfortunate than others.
>>53885821
ROPs are usually tightly tied together with L2 cache and memory controllers, since they tend to be the biggest bandwidth pigs.
Nvidia might not consider them parts of their Streaming Multiprocessor blocks, and there was no more detailed diagram of the GP100's memory controller stuff, so it's hard to tell.
>>53884585
it's funny because Nvidia screwed tomb raider even though it has some bad half baked dx12 implementation and Hitman is really good, makes Nvidia look like shit oh well.
>>53886091
>early access cancer
>being good
Pick one.
>>53886091
Don't act like hitman hasn't run much better on amd hardware over the years.
>>53885424
2016 Year of AMD
>>53885673
>>53886613
Because it doesn't have gimpworks.
>>53885424
>>53886797
God i hope so.
Zen and polaris is gonna be so comfy if it lives up to the hype. But there was bulldozer
>>53886959
because it was a console port.
>>53885871
And Nvidia's 3.5GB is 4GB of AMD's, right?
So I'm probably just really stupid and missing a detail here, but is the consumer stuff delayed until 2017 as well? I am in true dire need of a GPU upgrade (660) and have already extended myself to wait this long. Strongly considering just pulling the trigger on a 980 Ti.
>>53886959
As >>53887407 said, the game being a console port benefitted amd hardware greatly due to the amd hardware in the consoles. Many games this year have been like this.
>>53887791
go for AMD family
>>53887785
hawaii has no compression over the pci bus or in-memory compression, so yes, the 970's 3.5gb will actually hold more than AMD's 4gb. This was why AMD dropped the 4gb 290 entirely when they rebranded it.
>>53887829
I have a G-Sync monitor(it was the best option at the time) so I sort of have a vested interest in making use of that. I probably would otherwise.
>>53887834
Nvidia's DCC implementation doesn't compress in memory dude. AMD's does (ie on Fury) but it also compresses less comparatively.
>>53887834
Nvidia compress only to increase bandwidth, not the actual capacity.
>>53883756
>Using even more power
Good work nvidia
Meanwhile, AMD are making passive VR/4K cards.
>>53888032
AMD considers 20FPS at 4K payable.
>>53887936
>Nvidia's DCC implementation doesn't compress in memory dude.
>>53887976
>Nvidia compress only to increase bandwidth,
kek, clueless AMD shills who clearly don't know about DXTn
>>53888090
this too
https://en.wikipedia.org/wiki/Adaptive_Scalable_Texture_Compression
AMD hired the guy who developed this then decided not to implement it for some reason, meanwhile intel and nvidia have fully supported it for the last 4 years.
>>53888149
Is this the same as delta colour compression?
>>53888243
delta color compression is just for the transition from pcie bus to the gpu.
>>53888265
delta color compression is for buffers being rasterized to video RAM.
i.e., saving memory bandwidth when rendered pixels have low variance from their neighbors.
there are tons of texture compression formats, but DCC is not one of them.
>>53888090
>>53888149
>>53888265
Explain more about this. Is this why most nvidia cards come with less vram than their amd equivalents? How does this translate to a gaming experience? Will it cause more latency or LOD pop in or similar? And why don't amd use this if it's beneficial. Surely they can cut down the price of their cards if they implement these compression techniques and remove like 4gb of the 8gb vram on their 390 and 390x
>>53888346
DCC is for renderer output, so shaving maybe half the size off framebuffers.
The vast majority of memory is used to hold pools of art assets, not output buffers.
AMD did start using delta compression in the 285 IIRC, so it's not like they're totally ignoring the technology.
>>53888437
So what about this technique >>53888149
Won't these come with certain penalties like latency due to compression and decompression?
>new card is not intent to play games
>"muh muh it's not good for my gaymens"
>>53883756
>300W TDP
The housefires had returned
>>53887791
Same boat as you. Gsync monitor and need a GPU upgrade. This thread makes it seem like we should just buy 980 ti's and hold on these new GPUs.
>>53890356
Buy now, wait till 2018
>>53884410
Wasn't Pascal supposed to just be a die-shrink with some improvements?
>>53890356
Ditto, seems like the options are wait for polaris, or buy a 980ti
>>53884379
>970
>Run 1440p with steady and comfy 60 fps
>Current res 1050p
Medium jej
>>53885198
So GP100 is for people who want to do real work (video editing isn't real work), or aren't poor.
>>53890707
>play 1080p WoW and Dota 2 at a nice steady 144fps with GTX 970
Why would I bother upgrading?
This is the best thread for this question.
How much would the Asus strix 970 be worth in August?
>>53888866
Face it, no-one on here uses video cards for anything but MUH VIDYA.
>>53890856
The same. The new pascal cards are out after August I believe. Don't get a 970 anyway. Get atleast a 980 if you plan on keeping it for a few years. You never know what nvidia will do about that 3.5 gb problem because games are currently being limited to 3.5 via drivers but once they stop regular driver updates for the 9xx series you might run into problems with that.
I have a 980 myself and got it off ebay for 320 quid. It's oc to 980 ti~ stock levels so it's a damn good card. And you get that full 4gb ram.
>>53890976
I'm actually going to sell my asus 970 to get a new GPU. Will the 1080 (and the amd equivalent) be released by August?
>>53890812
>gtx 970
>for wow and dota2
>>53890993
I'm actually not 100% sure when any of the cards are being released. There's alot of speculation about that but I think the nvidia cards won't be out till after August but I do believe the amd cards will be out before August.
If you're unsure then just sell it now. I actually gave a colleague who was in a similar position to you the last time these cards came out a few years back. He just sold his card around march/April because he was unsure and focused his time on his studies and holidays. Before he knew it the new cards were out and he could start gayming again.
Tldr sell it now if you're unsure to get the highest resale price and occupy yourself with other shit till the new cards come out
>>53891093
>not having atleast a titan x for wow and dota
jfc
>>53891106
Well, I'm not a student or anything. I work at a paint/carpet/lino store (we sell stuff to decorate your walls and floor). So yeah I game a lot, I also work as a freelance web designer/it guy (come to you and fix your computer shit up for money), but I only get any work during holidays, and it's mostly people who know me/my previous clients. Only thing else I do in my free time is hang out with friends (read: get high), and try out every Linux distro + de combination.
>>53891233
Dino?
>>53891233
Yeah that's plenty of shit to be occupied with. Before you know it months will go buy and the new cards will be out. All the recent games are stale as fuck and aren't even worth playing and all the good games will be out near September except for ds3 which is most probably the only game worth playing now, but it can wait.
You can run all your mainstream mobas and shooters like LoL or cs:go on intel integrated graphics with playable framerates anyway.
>>53891093
>144fps
Honestly sounds like good news for me. I've been looking at monitors and all the ones that I prefer are Freesync not Gsync. Polaris is designed for 14nm right, not just a die shrink?
>>53883756
>>53885424
>yfw Raja is eating celebratory curry right now
>yfw kind of want
>>53891590
correct
Is pascal going to have 8.5 gb of vram? I really want it to, just to see the kind of shitstorm that would happen here/
Pascal such a disaster.
I am a #AMDmissile now.
#NeverNvidia
Nvidicucks lose again!
>>53885616
>mfw i have real pre production shitte since fermi/bulldozer/sandy bridge
Actually i have an am4 mobo with a high-ish end bristol ridge apu and whatever i say wint matter since i dont show proof or leaks because im not an idiot and i like my job lmao
>eetz gonna be gewd guise as powerfuk as a 980 ti guise believe me guise.
>>53892861
#gay
>>53885149
MR WORLDWAH
>>53885616
Ah yes cont. to >>53892998
Im more than sure that amd will make even low end stuff on 14nm lpp, also i bet they already has engineering samples from everything from low end notebook cpu's to server grade opterons, even their gpu line is already being tested since a lot.
>>53892809
7.5GB, benchmarks confirm
http://wccftech.com/nvidia-pascal-3dmark-11-entries-spotted/
>>53885424
and now the long wait for vega. i really wanted to sell my 980 ASAP but the only backup GPU i have is a 460, sucks. i imagine by Q1 2017 regular 980's will be only worth $250-300.
>>53893198
I'd be surprised if a 980 ti kept that price by then
I wonder how this compares to the S9300x2. Sure they are intended for different tasks but the S9300x2 is packing some serious compute.
ITS
HAPPENING
>>53893285
S9300x2 800 GFLOPS FP64 4GB HBM
P100 5.3 TFLOPS FP64 16GB HBM2
Not even a competitor, AMD is irrelevant
>buy pc without gpu waiting on pascal
>this shit happens
Any advice friends?
>>53893792
Just get Pascal/Polaris anyways. You've waited this long already.
P100 is a server GPU and doesn't even have a a PCI-e port. It's not a consumer GPU. Kill yourselves.
>>53893792
Just get the best GPU you can afford right now, waiting is a meme.
>>53893852
It's a meme when they're a year away not when they're literally 3 months away
>>53893705
>>53893718
What is?
>>53893997
Just like Zen was last year? Just like Polaris was in January? Unless you are absolutely certain that the cards are launching in the time frame you are willing to wait it's stupid to do so. Even if you wait the for the new cards and buy them it's still a gamble because you don't know if you are getting fucked like with the 970 or if you bought the Titan X and one month later the 980Ti comes out. And if you think prices are going to drop in with the next generation just check the 700 series prices.
When buying new tech, something newer is allways "around the corner".
>>53885424
Thanks dog.
>>53887834
>so yes, the 970's 3.5gb will actually hold more than AMD's 4gb
nice trolling m8
>>53894164
>Just like Zen was last year? Just like Polaris was in January?
Nobody ever said these things except a bunch of kids on /g/ who can't wait for a few fucking months so they overhype the shit out of everything and make some terrible guesses. Get your news from a source that isn't /g/.
JUST
>>53891272
Dino?
>>53891372
Well the thing is, I need a dp port for my 144hz 1440p monitor, which I won't have without a gpu. I do also play cs:go mostly.
Kys
>>53890507
no. it was supposed to be A BIG JUMP.
Seriously. Why try. Welcome to the new age, where AMD is just happy to be present. We had our fun with CPUs, not it's GPU turn.
>>53884263
What marketing, I haven't seen shit
It's over amd won. Pascal is nothing more than a maxwell refresh at 16 nm while amd is designing a new architecture with 14nm.
>>53886091
>DX12 turns out to really not be that great at all
>Two gmaes come out with DX12
>Have optional Nvidia effects
>Doesn't run well on really anything
"what the fuck nvidia, I blame gimpworks. I turned on these graphical effects that require more computation power and my framerate went down, fuck nvidia AMD FOR LIFE"
>>53886091
>DX12 not giving AMD cards +200fps with their ayysink shaders
>"It's all your fault Nvidia"
>>53898553
it gives at least 5% lead and up to 25% in proper optimized games
>>53899912
>Game doesn't run well on Nvidia cards
HAHAHA BTFO NVIDIA DEAD
>Game doesn't run well on AMD cards
Well you see the game needs proper optimization
>>53883756
12 years ago you would get your dick cutted to have GPU like this
Well, people seem to like it.
>>53902765
>market cap
> Nvidia 19.36b
> AMD 2.17 b
>Nvidia is literally worth 10 times more as a company than AMShit
>>53902879
Yet they struggle to make a consumer gpu that outperforms a 2 year old rebrand in the same price range.
>>53902765
Because us investors know that the money is in AI/HPC and NOT gaming. Gaming should be a side-business, if it should even bother still existing. Consoles are good enough for the vast majority of video game consumers. Discreet graphics cards for the gaming market is shrinking and deservedly so.
>>53902949
>struggle
>implying they're even trying
>>53902949
>Struggle
All they have to 'struggle' to do is make a new GPU that performs marginally better than AMShit
why do people keep saying "just remove dp units"
if it was that simple why didn't they do it before?
armchair chip designers btfo.
>>53904094
>if it was that simple why didn't they do it before?
They did.
>>53904318
they've used the same chip for tesla and geforce forever
selling two versions of a chip requires two tapeouts and thus doubles the cost, which is already in the hundreds of millions
>>53894113
New housefire powered rockets
The Way it's Meant to Burn
>>53898178
>I need a dp port for my 144hz 1440p monitor
just get a hdmi to dp adapter or buy a shitty $10 pcie display adapter of ebay
>>53904094
In Maxwell they could add them and remove them at will, they just needed the die space.
>>53903042
>>53903317
Exactly you cucks. Nvidia doesn't give a shit about you, they aren't gonna try hard to make something better because they don't have to. They have a monopoly and they know it. You're just gladly taking nvidias cock up your ass and spewing money all over the floor.
>>53904618
>nvidia doesn't give a shit
>still manages to make the superior and more popular products
>b-buy amd! they're shit but they care about you!
that's not how it goes, chump.
>>53904671
>superior
Lmao. For performance/price nvidia doesn't even compare. I can't believe you guys still shill them.
>>53904745
nigga the 390 and 970 are the most mentioned price/performance cards and the 970 is cheaper in my country. i can get a higher tier 970 like an evga ftw or gigabyte g1 edition for the same price as the most popular standard msi 390 AND those high oc on those 970 will beat out the 390.
>>53904844
high factory oc*
Is Intel/AMD the new hotness?
>>53904844
3.5GB
>>53904844
950 is still the price/performance king. AMD can't even win there unless you buy some sort of AMD APU/GPU combo for dirt cheap, and even there you have to deal with their shitty CPU.
>>53884159
Kek
>>53905076
Yes, and? It's not like that's actually an issue anymore. There's literally that one benchmark on AC:Unity which people like you keep spamming which has already been debunked.
>>53883756
>inb4 this is caused by technology reaching its physical limits
How fucked would we be?
>>53906239
It will be the end of the Lisp school and the return of good programming. Bad programming would be considered waste and environmentally unfriendly.
>>53906660
I'll take those demon trips as truth
>>53907118
newfag detected
>nice dubs btw
